You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
234<br />
categories including puffed snacks, potato chips,<br />
dried squid, popcorn, fish snacks, nuts, and prawn<br />
crackers. Before starting the first section, all<br />
panelists were advised how to evaluate samples.<br />
The 9-point hedonic scale was used for preference<br />
ratings (appearance, aroma, taste, texture, and<br />
overall liking), and there was a 10 minute break<br />
between each section.<br />
Data were then analyzed by SPSS 10.0.<br />
Cluster analysis was used to classify respondents<br />
by their overall liking scores for each product<br />
category. Overall liking of each respondent group<br />
for each product category was estimated from their<br />
attribute liking scores by multiple regression<br />
(stepwise) as the equation below.<br />
Overall liking = A (Appearance liking) + B<br />
(Aroma liking) + C (Taste<br />
liking) + D (Texture liking)<br />
1.2 Quantitative descriptive analysis (QDA)<br />
data<br />
Thirty semi-trained panelists were used<br />
to obtain QDA data for the same 21 snack products.<br />
They were selected and trained by the modified<br />
methods of ISO3972:1991 and ISO4120:1983,<br />
respectively.<br />
2. Quality function deployment<br />
A HOQ was built for each snack category<br />
to translate the attribute liking into QDA attributes.<br />
The importance scores of each attribute liking were<br />
considered from its coefficient in the multiple<br />
regression between the attribute liking scores and<br />
the overall liking scores.<br />
The impact between each pair of attribute<br />
liking and related QDA attribute, and the<br />
relationship between each pair of QDA attributes<br />
were considered by Pearson correlation. Cohen<br />
(1995) suggested that impacts in the middle of<br />
HOQ should be considered into 3 levels consisting<br />
of possibly related which was valued as 1,<br />
moderately related which was valued as 3, and<br />
Kasetsart J. (Nat. Sci.) 40(1)<br />
strongly related which was valued as 9. In the<br />
study, therefore, if a correlation coefficient was ≥<br />
0.9 or was an inverted U curve, then it was defined<br />
as being strongly related and scored as 9 for the<br />
impact of attribute liking and related QDA attribute<br />
(in the middle of house), and as + + or -- for the<br />
relationship between a pair of QDA attributes (on<br />
the roof). 0.6 ≤ correlation coefficient < 0.9 was<br />
defined as being moderately related and scored as<br />
3 and as + or –, respectively. In case of 0.4 ≤<br />
correlation coefficient < 0.6 it was defined as<br />
being slightly related and scored as 1. The<br />
contribution (priority) for each QDA attribute was<br />
then computed by summing the multiplication of<br />
impact and importance scores.<br />
The direction of goodness for each QDA<br />
attribute was considered from the sign + and – of<br />
the correlation coefficient; + meant high intensity<br />
is good and vice versa in case of minus. For the<br />
inverted U curve, it meant the intensity of that<br />
sensory attribute should be optimized.<br />
3. Reverse engineering<br />
3.1 Creating a set of equations<br />
The preference scores and QDA data<br />
were transformed to z-scores. The 21 products<br />
were then mapped in a coordinate system by<br />
reducing their QDA data with principal component<br />
analysis (PCA). The products’ PC scores and their<br />
square terms were used as independent variables<br />
in the equations for estimating the degree of<br />
preference for each of the sensory attribute as well<br />
as the overall liking, and the equations for<br />
evaluating the intensity of the QDA attributes by<br />
multiple regression (stepwise).<br />
3.2 Finding the target<br />
The overall liking equation was<br />
maximized for each product category by using the<br />
solver command in Excel 2000. These maximized<br />
PC scores were then used in the QDA attribute<br />
equations to get the target product profiles.